Anthropic CEO Dario Amodei identifies compute as key business dilemma

Dario Amodei has named the problem that keeps an AI lab’s finance team awake at night. The Anthropic CEO says compute is the company’s deepest business dilemma.

The issue is simple to describe and hard to solve. AI companies must commit to chips and data centers years before they know how much revenue those resources will produce.

A bet placed years in advance

Amodei laid out the tension at the New York Times DealBook Summit in December 2025. He called it a “genuine dilemma” in capacity planning.

Data centers and chip orders have long lead times. Demand for AI, by contrast, can swing sharply from one year to the next.

Amodei described two ways to get it wrong. Buy too little compute and customers may leave for competitors who can serve them.

Buy too much and the bills could become unsustainable. In the worst case, Amodei suggested, overcommitting could push a company toward bankruptcy.

The numbers behind the dilemma

Anthropic’s growth explains why the planning problem is so acute. The company’s revenue went from near zero in 2023 to $1 billion in 2024.

The company expected to reach an annualized run rate of between $8 billion and $10 billion by the end of 2025.

Later disclosures put dollar figures on the commitments. Anthropic’s S-1 filing, tied to IPO disclosures in November 2026, revealed $900.9 billion in long-term commitments, mostly related to compute and infrastructure.

Those obligations stretch through 2045. Of that total, $70.7 billion is due by September 30, 2027.

Compute costs also proved volatile as a share of revenue. In one quarter of 2026, compute expenses equaled 48% of revenue before the figure fluctuated.

Planning for the bad scenario, not the dream one

Amodei’s answer to the problem is a deliberately cautious one. He has argued for planning around the “10th percentile” scenario rather than the most optimistic outcome.

The idea is to size commitments so the company survives even if growth comes in well below expectations. The goal is avoiding insolvency. Amodei has framed this approach as a contrast with competitors who are pursuing more aggressive buildouts.

Even so, Anthropic is not exactly sitting still. The company has continued expanding compute rapidly.

Why this matters beyond Anthropic

Every major AI developer faces the same timing mismatch between infrastructure spending and revenue.

For investors, Anthropic’s revenue trajectory is striking, but long-dated compute commitments reduce flexibility if the market shifts. A company locked into infrastructure obligations through 2045 cannot easily scale back if AI demand cools.

The 48% compute-to-revenue figure in one quarter is worth watching closely. The near-term marker is the $70.7 billion due by September 30, 2027.

Amodei’s framing amounts to a public admission that nobody really knows how big AI demand will get. For a company carrying $900.9 billion in commitments, the forecast matters a great deal.